[ntp:questions] ntpd: time reset problem

David Lord snews at lordynet.org
Sat Sep 26 17:58:20 UTC 2009


Unruh wrote:
> David Lord <snews at lordynet.org> writes:
> 
>> David Lord wrote:
> 
>>> Below are values for offset and jitter from polls at
>>> 30 min intervals.
>>>
>>>           BR-304         gps18x-lvc         gps18x-lvc
>>>              rmc                rmc            rmc+pps
>>>    polls  offset      polls  offset      polls  offset
>>>   %     n   (ms)     %     n   (ms)     %     n   (ms)
>>>  50    90   9.23    51    27   0.07    54    76  0.002
>>>  95   170  53.02    92    49   0.16    99   140  0.005
>>>  99.5 178 435.84   100    53   0.20   100   141  0.006
>>>
>>> Problem with BR-304 was lack of sensitivity and being
>>> unable to keep sufficient satellites in view, even
>>> though positioned with slightly better view of sky than
>>> the Garmin.
> 
>> Oops again since I'd deleted jitter results as I'd missed sorting
>> them separate from the offsets. I'll put up a link to full table
>> of results, once I've finished trying various sources and also
>> settled on a reasonably consistent method for showing results.
>> These can even include those from when using chrony with
>> demand dial.
> 
>> Anyway here are what I have so far:
> 
>> Source Polls   % offset jitter   % offset jitter   % offset jitter
>>       @30min       (ms)   (ms)       (ms)   (ms)       (ms)   (ms)
>> (1)      141  54  0.002         99  0.005         100 0.006
>> (2)      100  51  0.07          92  0.16          100 0.20
>> (3)      179  50  9.23          95  53.02        99.5 435.8
>> (4)      203  50  0.185  0.086  95  0.642  0.255 99.5 1.111  0.488
>> (5)       53  50  0.165  0.070  95  0.621  0.246 99.5 0.673  0.333
>> (6)      164  50  1.928         95  9.162          99 21.48
>> (7)      174  50  0.847         95  2.548          99 6.548
> 
> Your number 3 is really terrible. That gps should not be delivering the
> time if it cannot lock on to enough sattelites. To deliver times that
> are 1/2 sec out is really pretty terrible. 
> 
> 
>> VIA EPIA C3-600
>> NetBSD 4.1/4.99x/5.0 then 5.0.1
>> (1) GPS_NMEA    gps18x-lvc + rmc + pps
>> (2) GPS_NMEA    gps18x-lvc + rmc
> 
> What is rmc?

NMEA output is a whole bunch of sentences giving different data
and these can be split over more than a second (at default bps).

RMC is just one of the outputs that gives sufficient data for
the driver so setting the gps to only give minimal output
might be helpful (or not). The gps needs command string to
set required output sentences. On rollover any that cant fit
in a single second are split over two or more seconds.

RMC = recommended minimum
GSV = satellites in view

.....

> 
>> I didn't need serial-usb until very recently as all pcs here
>> except new netbook have serial port so a converter wasn't
>> needed when I tried out the GPS modules. I can't get converters
>> to work reliably from ttl out of either Conrad module or my
>> PPS extractor monostable both used for (5) above. I suspect a
>> ttl to rs232 conversion is needed or some simple hack.

> The problem is the interrupt. There is no way to deliver an interrupt to
> usb AFAIK. From what I understant, seems that there is random latency in the
>  usb system as well.

Thanks I'd already got that pps was probably out from your
previous post. I can't even get data through the link though but
I should have mentioned a standalone modem is seen ok, but I've
not tried a dialout, just AT commands. From using mobile
broadband over usb with ntpd I can see there is a problem with
delays but I'm happy to get some numbers from radioclocks for
comparison. My MSF setup has sufficient outputs to steal from
and generate -5V for a better rs232 output if that's what's
needed.

Off topic even more, I'd setup my PPS extractor using sig-gen
and 'scope but when tried it was way off (aliasing?). The
radioclkd2 debug mode made setup of 950ms pulses very easy.
My old 'scope had 4sec long persistence phosphor and many
seconds timebase which would also have been a big help now,
but 4kV and 1.5kV supplies to crt and 300V for line output
valves all naked on bench are probably not allowed now :-(

     Jonathan Buzzard
<http://www.buzzard.org.uk/jonathan/radioclock.html>

     and Jon Atkins
<http://www.jonatkins.com/page/software/radioclkd2>


David




More information about the questions mailing list